基于多特征融合的运动想象脑电信号分类研究

Research on Motor Imagery EEG Signal Classification on Multi-Features Fusion

针对脑机接口(BCI)系统中常用的运动想象脑电信号,提出多特征融合的方法,提高其分类准确率。采用自回归(AR)模型、小波包变换、共空间模式(CSP)对BCI2003竞赛数据进行信号处理,提取AR模型系数、小波包熵差、方差作为特征,并在分类器上测试准确率。实验结果表明:多特征融合的分类准确率均高于单特征,保持在90%左右,可以更好地表征运动想象脑电信号。数据处理所选取的时间段在4-7s,实验结果具有较好的鲁棒性和泛化能力。

Proposes a method of multi-features fusion to improve the classification accuracy of motor imagery electroencephalogram signals commonly used in brain computer interface(BCI) system. Uses the autoregressive(AR) model, wavelet packet transform and common spatial pattern(CSP) to process BCI2003 competition data. AR model coefficient, extracts the wavelet package entropy difference and variance as the characteristic, and tests the accuracy on the classifier. The experimental results show that the classification accuracy of the multi-features fusion is higher than that of the single feature. It is maintained at about 90%, which can represent the motor imagery electroencephalogram signals better than the single feature. The time period of data processing is 4-7 s, and the experimental results have good robustness and generalization ability.